JPH1165599A - Audio compression / expansion method and apparatus, and storage medium for storing audio compression / expansion processing program - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To enable efficient and high-quality voice compression / expansion with simple processing and to be advantageous for hardware and parallel processing. SOLUTION: A speech segment clipping unit 1 clips a speech segment in a predetermined section as a speech segment to be processed, and a similarity judgment unit 3 judges a similarity by referring to a speech segment table 4 including a plurality of types of speech segment groups. The voice segment selecting unit 5 selects a voice segment having a part with the highest similarity. Then, based on the data about the speech segment selected by the encoding unit 6, the speech segment to be processed is encoded. Further, after performing the decompression process or after extracting the spectrum envelope parameters, the speech unit updating unit 11 updates the content of each speech segment stored in the speech segment table.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio compression / expansion method and apparatus for efficiently compressing / expanding audio signals by simple processing, and a storage medium storing an audio compression / expansion processing program.

[0002]

2. Description of the Related Art Various methods have been proposed as encoding methods for compressing and expanding audio signals. Part 1
For example, there is JP-A-59-116973 (hereinafter referred to as a first prior art).

The first prior art is a means for dividing input audio data for each short time to obtain a short-time audio signal sequence.
A spectrum envelope parameter extracting means for extracting a spectrum envelope parameter from the short-time speech signal sequence, an impulse response sequence calculating means for calculating an impulse response sequence based on the spectrum envelope parameter, and an autocorrelation function using the impulse response sequence Means to calculate columns,
Means for calculating a cross-correlation function sequence using the impulse response sequence and the short-time audio signal sequence, means for calculating and encoding a drive excitation signal sequence using the auto-correlation function sequence and the cross-correlation function sequence, a spectral envelope code And a means for combining and outputting a driving sound source signal, and a target signal calculating means for applying a predetermined correction to the short-time audio signal.

According to the first conventional technique, the position and the gain of the drive excitation pulse can be efficiently determined when encoding the speech, and the amount of calculation and the amount of memory used can be reduced. Some effect can be obtained.

However, according to the first prior art, when a speech signal such as a female voice is encoded and then speech synthesis is performed, a large number of driving sound source pulses are extracted in order to obtain high-quality speech synthesis. Because of the necessity, there is a problem that the compression ratio is deteriorated.

That is, a female voice is more complicated than a male voice, and it is necessary to extract a large number of driving sound source pulses in order to obtain a highly accurate synthesized sound. Would.

On the other hand, as techniques for obtaining a high compression ratio, there are JP-A-63-37399 (hereinafter referred to as a second prior art) and JP-A-3-4300 (hereinafter referred to as a third prior art).

In the second prior art, pitch estimation is performed from a speech signal, a residual between an estimated value from a past pulse train and an actual signal is obtained, and a driving excitation pulse is calculated based on the residual. is there.

In the third prior art, pitch estimation is performed, and a driving sound source (multi-pulse) for one pitch section is estimated. Then, by correcting the gain and phase of the multi-pulse, the other pitch section is approximated by correcting the other pitch section. Further, the second multi-pulse is estimated from the residual between the estimated value and the actual value. Note that a noise codebook may be used in addition to the multi-pulse signal.

[0010]

According to the second and third prior arts described above, a cycle in which the same waveform is repeated is obtained, a next cycle is estimated from a previous cycle, and the estimated portion is compared with an actual voice. Since the difference from the waveform is calculated and the driving sound source is calculated based on the difference, a high compression ratio can be realized.

However, there is a problem that the amount of calculation is large because it is necessary to obtain a pitch and a difference, and a large-capacity memory is required to store such data.

In addition, since a residual is obtained and a driving excitation pulse is calculated based on the residual, when a part of data is lost, the lost data has a great effect on subsequent calculations. However, there is a big problem that high-precision speech synthesis cannot be performed.

As described above, each of the conventional techniques has various problems. For example, the first prior art is a basic technique for obtaining a driving sound source pulse. However, in order to increase the quality of a synthesized sound, it is necessary to make many driving sound source pulses, and it is necessary to generate many driving sound source pulses. In particular, there is a problem that the compression ratio is deteriorated with respect to a simple audio data. Further, the second and third prior arts provide high compression ratios, but have a problem of a large amount of calculation and a large amount of memory used, and furthermore are vulnerable to data loss due to the use of difference information. There's a problem.

In recent years, portable information devices for handling audio data have been used in a wide range of fields. In this type of portable information device, since the calculation speed and the memory capacity of the CPU are greatly restricted, it is a serious problem that the calculation amount and the used memory amount are large. In addition, the method using difference information has many problems in terms of improving the performance of products in information devices that need to consider the lack of data, and is not limited to mobile devices, and is not limited to real-time transmission over computer networks. Is also greatly affected by the transmitted data.

As described above, each of the conventional speech coding methods has in common that the processing is complicated.
There is a problem that it is relatively difficult to increase the speed by hardware and parallel processing. In particular, the processing including the processing for obtaining the pitch period requires a large amount of calculation and has a large influence when an error occurs. Further, the conventional method using the impulse response based on the spectral envelope parameter and the driving pulse has a problem that discontinuity occurs before and after the pulse, which appears as noise.

Accordingly, the present invention provides a simple processing content, enables easy hardware and parallel processing, enables efficient encoding, and compresses audio data at a relatively high compression rate. It is an object of the present invention to provide an audio compression / expansion method and apparatus, and a storage medium storing an audio compression / expansion processing program.

[0017]

According to a first aspect of the present invention, there is provided a voice compression / expansion method for extracting a voice segment of a predetermined section from an input voice as a voice segment to be processed, and including a plurality of types of voice segment groups. Referring to the table, by comparing the similarity between each speech segment in the speech segment table and the speech segment to be processed,
It is characterized by including a process of selecting a speech segment having the highest similarity and encoding the speech segment to be processed based on data on the selected speech segment to create encoded data.

According to a second aspect of the present invention, in the first aspect of the present invention, after the encoded data is created, the encoded data is decompressed, and the decompressed data is subtracted from the processing target speech piece to obtain a residual. For the obtained residual waveform, a process of comparing the similarity between each of the voice segments in the voice segment table and the residual waveform by referring to the voice segment table including the plurality of types of voice segment groups is described. More than once, encoded data is obtained.

According to a third aspect of the present invention, in the first or second aspect of the present invention, the speech segment stored in the speech segment table is already subjected to compression / expansion processing which is temporally backward from the speech segment to be processed. Speech piece created using the speech waveform
At least a speech fragment created using the temporally forward predicted speech waveform and the temporally backward predicted speech waveform estimated by the spectral envelope parameter, and a speech fragment created by a noise component, each of which is encoded The content is updated after the expanded data or after extracting the spectral envelope parameters.

[0020] Further, the invention of claim 4 provides the invention according to claims 1 to 3.
In any one of the inventions, each of the voice segments has a section that is temporally longer than the voice segment to be processed, and when determining the similarity with the voice segment to be processed, the range of the length of each voice segment is determined. In, the similarity to the speech segment to be processed is determined, and the speech segment having the part with the highest similarity is selected.

According to a fifth aspect of the present invention, in the fourth aspect of the present invention, the coded data represents a voice segment number having a portion having the highest similarity and a portion within the voice segment. It is data represented by position data and parameters for amplitude adjustment, and further data to which a spectrum envelope parameter is added as necessary.

According to a sixth aspect of the present invention, there is provided a voice compression / expansion apparatus according to the present invention, wherein a voice segment extraction unit for extracting a voice segment of a predetermined section set in advance from an input voice as a voice segment to be processed, and a spectral envelope from the input voice. A spectral envelope parameter extraction unit for extracting parameters, a speech unit table storing a plurality of types of speech units, and referring to the speech unit table,
A similarity determining unit that determines the similarity by comparing the similarity between each of the voice segments in the voice segment table and the target voice segment, and based on the similarity determined by the similarity determining unit, A speech segment selection unit for selecting a high speech segment, an encoding unit for encoding the processing object speech segment based on data on the speech segment selected by the speech segment selection unit, and encoding by the encoding unit And output the encoded data as encoded data, and in some cases, create encoded data obtained by adding the spectrum envelope parameter extracted by the spectrum envelope parameter extraction unit to the data encoded by the encoding unit. And a coded data output unit to be output.

According to a seventh aspect of the present invention, in addition to the above, a decompression unit for decompressing the data encoded by the encoding unit, and subtracting the data decompressed by the decompression unit from the speech piece to be processed. A residual generation unit for obtaining a residual, and updating of the content of the speech unit stored in the speech unit table using the data expanded by the expansion unit or the spectrum envelope parameter extracted by the spectrum envelope parameter extraction unit. And a voice segment updating unit for performing the process.

According to an eighth aspect of the present invention, in the seventh aspect of the present invention, the similarity determination section, speech piece selection section, encoding section, decompression section, and residual generation section form a loop in a processing procedure. For the residual waveform obtained by performing similarity determination, speech segment selection, encoding, decompression, and residual generation processing, the speech segment table is referred to, and each speech segment in the speech segment table is referred to. And comparing the similarity between the residual waveform and
After performing one or more times, encoded data is created and output.

Further, the invention of claim 9 is the invention of claims 6 to 8
In any one of the inventions described above, the speech segment stored in the speech segment table is a speech segment created using an already-compressed and decompressed speech waveform temporally later than the speech segment to be processed, and a spectral envelope. At least a speech segment created using the temporally predicted speech waveform and the temporally predicted speech waveform estimated by the parameter, and a speech segment created by a noise component, and each speech segment is subjected to the speech update processing. The content is updated by the section after the expansion process or after the extraction of the spectral envelope parameter.

According to a tenth aspect of the present invention, in any one of the sixth to ninth aspects, each of the voice segments has a section longer in time than the processing target voice segment. At the time of similarity determination, the similarity to the processing target voice segment is determined within the range of the length of each voice segment, and the voice segment having the portion with the highest similarity is selected.

According to an eleventh aspect of the present invention, in the tenth aspect of the present invention, the coded data indicates a voice segment number having the portion having the highest similarity and which portion in the voice segment. It is data represented by position data and parameters for amplitude adjustment, and further data to which a spectrum envelope parameter is added as necessary.

Further, according to a twelfth aspect of the present invention, there is provided a storage medium storing a voice compression / expansion processing program, wherein the voice compression / expansion processing program cuts out a voice segment of a predetermined section from an input voice as a voice fragment to be processed. Referring to the voice segment table including the type of voice segment group, comparing the similarity between each voice segment in the voice segment table and the processing target voice segment, selecting the voice segment having the highest similarity, and selecting Encoding the processing target speech piece based on the data about the speech piece,
Depending on the case, while performing processing of creating encoded data also including the spectral envelope parameter, and after the expansion processing of the encoded data or after extracting the spectrum envelope parameter, each stored in the speech piece table To update the content of the voice segment.

As described above, according to the present invention, the similarity between each speech segment in the speech segment table and a speech segment to be processed (for example, a speech segment having a length of about 4 msec) extracted from the input speech is compared. As a basic process, a speech segment having the highest similarity is selected, and the speech segment to be processed is encoded based on data on the selected speech segment. As a result, the encoding can be performed by a very simple process, which can be advantageous when performing hardware processing and parallel processing.

After the encoded data has been created, the encoded data is expanded, the expanded data is subtracted from the processing target audio fragment, and the residual generation process is performed. By referring to the table and performing a process of obtaining similarity at least once to obtain encoded data, encoded data with higher precision can be obtained.

The speech segment stored in the speech segment table is estimated by a speech segment created using an already-compressed and expanded speech waveform temporally behind the speech segment to be processed, and a spectral envelope parameter. By having at least a speech segment created using the temporally forward predicted speech waveform and the temporally backward predicted speech waveform, and a speech segment created by a noise component, efficient and high-quality encoding of the input speech is achieved. Accurate encoding becomes possible. In particular, when using a predicted speech waveform estimated by the spectrum envelope parameter,
Conventionally, temporally predicted speech waveform (impulse response)
Although it is common to use only a speech segment, the present invention creates a speech segment using a temporally forward predicted speech waveform and a temporally backward predicted speech waveform estimated by a spectral envelope parameter.

Thus, in addition to the forward predicted speech waveform,
The use of a backward predicted speech waveform that is temporally backward has the effect of reducing noise. That is, if a speech piece using only the impulse response (forward predicted speech waveform) is used, the speech piece becomes a speech piece whose waveform rapidly rises from a state where the speech level is almost 0, and is compressed and decompressed using the speech piece. When processing is performed, there is a problem that a discontinuous point is generated and the part appears as noise. On the other hand, if a backward predicted speech waveform temporally backward is used, discontinuities can be reduced as much as possible, and the quality of compressed and expanded speech can be greatly improved.

Further, the content of each speech piece is updated after the expansion processing of the encoded data or after the extraction of the spectral envelope parameter, so that
Unlike a conventional codebook having a fixed content as in the related art, an optimum speech segment is always stored for a speech segment to be processed, and high-quality encoding can be performed.

The encoded data includes a voice segment number having a similar partial voice segment, position data indicating which portion in the voice segment, and data expressed by an amplitude adjustment parameter. In some cases, it can be represented by data to which a spectral envelope parameter is also added. Therefore, the encoded data becomes data of about several bytes, and significant data compression becomes possible. In general,
Since the voice does not change abruptly, the spectral envelope parameter changes slowly when each of the voice segments to be processed is assumed to be about 4 msec, and the frequency of the spectral envelope parameter is about once every 10 voice segments to be processed. Sufficient accuracy can be obtained by extracting the envelope parameters. Therefore, even if the spectral envelope parameters are added, the data can be significantly compressed.

[0035]

Embodiments of the present invention will be described below. Before describing a specific embodiment, first, basic processing contents of the embodiment of the present invention will be described.

FIG. 1 shows an input speech waveform. From such an input speech waveform, a speech piece of, for example, about 4 msec is cut out. The cut-out speech segment (hereinafter, referred to as a speech segment to be processed) h1 is compared with the speech segment stored in the speech segment table, and the speech segment having the highest similarity is selected from the speech segment table. Then, encoded data is created using the speech piece. Note that the processing target speech piece was set to 4 msec in the system used in this embodiment.
This is because the best result can be obtained by cutting out with a length of about msec. In other words, the length of the voice segment to be processed is 4 msec.
When the length is shorter than the above, the sound quality is improved, but the compression ratio is reduced. When the length is longer than 4 msec, the compression ratio is advantageous, but the sound quality may be deteriorated. .

The speech segment table referred to here has speech segments (four speech segments A1 to A4 in this example) created from a plurality of elements as shown in FIG. The method of creating the voice segment will be described later. Note that the latest voice segment is always stored in the voice segment table, and the voice segment table shown in FIG. 2 shows the contents of the voice segment table at a certain time.

Now, assuming that the speech segment table shown in FIG. 2 has the latest contents, in FIG. 1, the cut-out speech segment h1 of about 4 msec is replaced with any speech segment in the speech segment table. Determine which part is most similar. In this case, the speech segment h1 to be processed is determined to have the most similar portion from the position p1 of the speech segment A2 in the speech segment table. The most similar part will be referred to as a similar part.

Thus, the encoded data of the speech segment h1 to be processed can be represented by the speech segment number A2, the position p1, and the magnification for matching the speech level in the speech segment table.

That is, in this case, since there are four voice segment numbers A1 to A4 in the voice segment table in this case, they can be represented by 2 bits, and the position p1 corresponds to a case where the length of each voice segment is 16 msec. For example, since it is 128 sampling points (assuming that the sampling frequency is 8 kHz), it can be represented by 7 bits. Also, if the audio level is adjusted in 128 steps, for example, it can also be represented by 7 bits. Therefore,
When these are summed up, they can be expressed as 16 bits, that is, 2 bytes of data.

On the other hand, if the number of sampling points is 32 and the number of sampling points is 32, the speech segment h1 to be processed has a data amount of about 2 bytes at each sampling point.
There will be a data amount of 4 bytes. Therefore, the data amount after encoding is 1 /
It becomes 32.

When the spectral envelope parameter is used, about 4.5 bytes are required as the data. However, in general, the voice does not change abruptly. Therefore, when it is considered that each of the voice segments to be processed is about 4 msec, the change of the spectrum envelope parameter is gradual, and about once every ten voice segments to be processed. Sufficient accuracy can be obtained by extracting the spectrum envelope parameter at the frequency of. Therefore, even if the spectrum envelope parameter is added, the encoded data can be significantly compressed from the original data. it can.

As described above, according to the present invention, the processing itself is simple, and efficient compression of audio data is possible.

Next, a specific embodiment of the present invention will be described.

FIG. 3 is a flowchart for explaining the processing procedure of the embodiment of the present invention. In FIG. 3, first, a processing target speech piece h1 of about 4 msec is cut out from the input speech (step s1). This process is the process described with reference to FIG. Then, it is determined whether or not to extract the spectrum envelope parameter (step s2). If the spectrum envelope parameter is required, the spectrum envelope parameter is extracted (step s3). Note that, as described above, since the voice rarely changes abruptly, the spectral envelope parameter changes slowly when each of the target voice segments to be cut is assumed to be about 4 msec. Therefore, sufficient accuracy can be obtained by extracting the spectral envelope parameter at a frequency of about once in 10 speech segments to be processed.

Then, in the next step s4, the speech segment having the highest similarity is selected with reference to the speech segment table at that time. For example, assuming that the content of the voice segment table at that time is the content shown in FIG. 2 with respect to the voice segment h1 to be processed at a certain point in time, the processing target voice segment h1 is the position of the voice segment A2 in the voice segment table. It is determined that the part from p1 is the most similar, and the voice segment A2
Is selected as a speech piece having a similar part.

Next, an encoding process is performed based on the data (speech segment number, position, magnification for matching the speech level) of the selected speech segment A2 (step s).
5).

Then, it is determined whether or not the compression processing has been completed (step s6). If the compression processing has been completed, the coded data that has been coded in step s5 is output (step s7). It is determined whether or not all compression processing has been completed for the input voice (step s8). If the processing has been completed, the processing is terminated. If not completed, the processing returns to step s1.

On the other hand, if the compression processing is not completed in step s6, the decompression processing (step s9) and the residual generation processing (step s10) are performed, and then the processing returns to step s4, and the processing is performed in steps s4 to s10. Loop processing is performed. Hereinafter, this loop processing will be described.

As described above, for example, it is determined that the portion from the position p1 of the speech segment A2 in the speech segment table to the speech segment h1 to be processed is the most similar, and the speech segment A2 having the similar portion is determined. Assume that it is selected. Then, the encoding process is performed based on the data (speech segment number, position, magnification for matching the speech level) of the selected speech segment A2, and the like. At this stage, the same processing is repeated several times without terminating the compression processing. That is, after the data is encoded in step s5, the encoded data is temporarily expanded (step s7), and thereafter, a residual generation process is performed (step 8).

This residual generation processing is performed by converting the encoded and expanded audio data into the original input audio (in this case,
In this process, the difference is subtracted from the processing target speech piece h1) to obtain the difference. That is, as shown in FIG. 4, the decompressed audio data H1 is subtracted from the processing target audio piece h1, and the residual d1 is obtained. Then, with respect to the obtained residual d1, a process of selecting a voice segment having a portion (similar portion) having the highest similarity with reference to the voice segment table at that time is performed. By performing such processing at least once, compressed data with higher precision can be obtained.
Sufficient accuracy can be obtained even about times.

The decompression process performed in step s9 is performed according to the processing procedure shown in the flowchart of FIG.

That is, the encoded data is input (step 11), and it is determined whether or not the spectrum envelope parameter is updated (step s12). That is, when the spectrum envelope parameter has been extracted, the value of the previous spectrum envelope parameter is updated to a new value of the spectrum envelope parameter (step s13).

Next, by referring to the speech piece table at that time, a speech piece having a part (similar part) having the highest similarity is selected based on the encoded data (step s1).
4). Then, decompressed data is created based on the selected speech piece data (step s15). Then, it is determined whether or not the processing has been completed (step s16). If the processing is not completed, the contents of the speech segment table up to that time are updated with the new speech segment using the data expanded in step s15 (step s17).

Then, if encoded data exists,
A similar process is performed on the encoded data.

Note that this decompression processing is used not only as one of the processing procedures in FIG. 3 but also in the decompression processing alone. For example, when coded data is stored in a predetermined memory, it is also used when decompressing the coded data.

When the decompression process is completed as described above, a residual is generated in the flowchart of FIG. 3 (step s10). That is, as described above, as shown in FIG. 4, the decompressed voice data H1 is subtracted from the voice piece h1, and the residual d1 is obtained. Then, with respect to the obtained residual d1, the speech segment having the highest similarity (similar portion) is selected with reference to the speech segment table at that time (the speech segment table newly updated after the decompression process). Is performed. By performing such processing at least once, compressed data with higher precision can be obtained. However, as described above, sufficient accuracy can be obtained even with about two times.

By the way, the speech piece table used in the above processing includes at least speech pieces created by the following elements.

(1) At present, speech data which has been already compressed and decompressed for the cut-out speech segment to be processed (speech data which has been compressed and decompressed later in time with respect to the speech segment to be processed) is used. Here, the expression that the time that has already passed is temporally backward and the time that is future is temporally forward are used.

For example, assuming that the input voice is as shown in FIG. 6A, the input voice up to a certain time t1 has already been subjected to compression / expansion processing, and the compressed / expanded voice waveform is as shown in FIG. 6B. I do. Assuming that the current speech segment to be processed is h1, a predetermined portion of the compressed and expanded speech waveform shown in FIG. Is used as a speech piece. This corresponds to, for example, the speech piece A2 in the speech piece table shown in FIG. The time length of the voice piece is about 16 msec.

(2) A temporally forward predicted speech waveform estimated from a spectral envelope parameter near a speech segment to be processed and a temporally backward predicted speech waveform that follows the waveform are used.

As described above, the spectral envelope parameter does not need to be sent for each extracted speech piece.
This is because the voice is considered to hardly change drastically. For example, if the spectral envelope parameter is sent at a rate of once for several to several tens of speech pieces to be processed, Good. In this sense, the expression “spectral envelope parameter near” the speech segment to be processed is used here.

As shown in FIG. 7, the temporally forward predicted speech waveform estimated from the spectral envelope parameter in the vicinity of the currently processed speech segment and the temporally backward predicted speech waveform continuous with it are shown in FIG. (Forward predicted speech waveform) In addition to x1, it indicates a temporally backward predicted speech waveform x2.

As described above, using the backward predicted speech waveform temporally backward in addition to the impulse response (forward predicted speech waveform) has an effect of reducing noise. That is,
If a speech piece using only the impulse response (forward predicted speech waveform) is used, the speech level suddenly rises from a state where the speech level is almost 0, and the compression / expansion process is performed using the speech piece. There is a problem that a discontinuous point sometimes appears as noise when the discontinuous point occurs. On the other hand, if a backward predicted speech waveform temporally backward is used, discontinuities can be reduced as much as possible, and the quality of compressed and expanded speech can be greatly improved.

(3) Use a noise waveform.

The noise waveform may be given by random numbers, or may be sampled from actual input speech.

As described above, the contents of the speech piece table used in the present invention include at least the speech pieces described in (1) to (3). And each of these speech pieces is 1
For example, a voice piece having a length of about 6 msec is held in a state as shown in FIG. 2, and the latest data is always stored.

FIG. 8 is a block diagram showing the configuration of the audio compression / decompression device of the present invention. In FIG. 8, the voice input from the voice input unit 1 is cut out by the voice cutout unit 2 as a processing target voice piece of, for example, about 4 msec as described above. The extracted speech segment to be processed is compared with some speech segments a1, a2,..., An in the speech segment table 4 by the similarity determination unit 3 to obtain a similarity. Then, the voice segment selection unit 5 selects a voice segment having a portion with the highest similarity (similar portion).

The encoding section 6 performs an encoding process based on the data (speech segment number, position, magnification for matching the speech level) of the selected speech segment. If the encoding process is terminated at this stage, the encoded data is output from the encoded data output unit 7. At this time, when using the spectrum envelope parameter, an encoding process is performed by adding the spectrum envelope parameter extracted by the spectrum envelope parameter extraction unit 8.

On the other hand, if the encoding process has not been completed, the encoded data encoded by the encoding unit 7 is expanded by the expansion unit 9, and the residual generation unit 10 performs the residual generation process. The decompression processing and the residual generation processing are the processing of steps s9 and s10 of the flowchart in FIG.

As described above, this residual generation processing is performed by subtracting the coded and expanded voice data from the original input voice (in this case, the voice segment h1 to be processed).
This is the process of taking the difference. That is, as shown in FIG. 4, the decompressed voice data H1 is subtracted from the voice piece h1, and the residual d1 is obtained. Then, with respect to the obtained residual, a process of selecting a similar partial speech segment having the highest similarity with reference to the speech segment table at that time is performed.

The decompression process performed by the decompression unit 9 is performed according to the processing procedure shown in the flowchart of FIG. Then, the speech piece updating unit 11 updates the contents of the speech piece table 4 using the decompressed speech data. Further, when the spectrum envelope parameter is extracted from the spectrum envelope parameter extraction unit 8, the speech piece updating unit 11 performs temporally forward predicted speech waveform estimated by the spectrum envelope parameter and temporally backward predicted continuous with it. Also update the audio waveform. In this way, the contents of the voice segment table 4 always store the latest voice segments.

Since the overall operation of the audio compression / decompression device having such a configuration has been described with reference to the flowchart of FIG. 4, the description of the operation will be omitted here.

The present invention is not limited to the above-described embodiment, but can be variously modified without departing from the gist of the present invention. For example, the processing target speech piece to be cut out is 4 msec in the above-described embodiment.
The reason is that the best result was obtained with 4 msec in the system used in the above embodiment. However, the numerical value may be different depending on the system to be used and the like, and the present invention is not limited to this, and an optimal time can be set according to the system to which the present invention is applied. Further, the contents of the speech piece table shown in FIG. 2 are an example, and the present invention is not limited to this.

Further, the processing program for performing the above-described audio compression / decompression processing of the present invention can be stored in a storage medium such as a floppy disk, an optical disk, or a hard disk. And a method of obtaining data from a network.

[0076]

As described above, according to the present invention,
Compare the similarity of each speech segment in the speech segment table with the speech segment to be processed cut out from the input speech, select the speech segment with the highest similarity, and based on the data for the selected speech segment. The processing of encoding the processing target speech piece is performed as basic processing. As a result, encoding can be performed with a very simple process.

After the coded data is created, the coded data is expanded, and the residual waveform obtained by subtracting the expanded data from the processing target audio piece is again converted to the audio signal. By referring to the table and performing the process of obtaining similarity a plurality of times to obtain encoded data, encoded data with higher quality can be obtained.

The speech segments stored in the speech segment table are estimated based on the speech segments created using the already-compressed and expanded speech waveforms temporally behind the speech segment to be processed, and the spectral envelope parameters. By having at least a speech segment created using the temporally forward predicted speech waveform and the temporally backward predicted speech waveform, and a speech segment created by a noise component, efficient and high-quality encoding of the input speech is achieved. High quality encoding is possible. In particular, when a speech segment is created from a predicted speech waveform estimated by the spectrum envelope parameter, in the present invention, in addition to the temporally forward predicted speech waveform estimated by the spectrum envelope parameter, the backward predicted speech waveform temporally backward Is used, noise can be reduced, and the quality of voice can be greatly improved.

Further, the content of each speech segment is updated after the expansion processing of the encoded data or after the extraction of the spectral envelope parameter.
Unlike a conventional codebook having a fixed content as in the related art, an optimum speech segment is always stored for a speech segment to be processed, and high-quality encoding can be performed.

The coded data includes a voice segment number having a similar partial voice segment, position data indicating a portion within the voice segment, and data expressed by an amplitude adjustment parameter. In some cases, it can be represented by data to which a spectral envelope parameter is also added, and significant data compression is possible.

As described above, the present invention makes it possible to efficiently and efficiently perform high-quality voice compression / decompression with simple processing contents, and can be made extremely advantageous when implementing hardware or parallel processing.

[Brief description of the drawings]

FIG. 1 is a diagram showing an example in which input speech is cut out in a predetermined section in order to explain an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a speech piece table according to the embodiment of the present invention.

FIG. 3 is a flowchart illustrating a processing procedure according to the embodiment of the invention.

FIG. 4 is a view for explaining processing for obtaining a residual component in the embodiment of the present invention.

FIG. 5 is a flowchart illustrating a decompression processing procedure according to the embodiment of the present invention.

FIG. 6 is an exemplary view for explaining an example of creating a speech segment in a speech segment table from a speech waveform after decompression processing according to the embodiment of the present invention;

FIG. 7 is a diagram illustrating an example in which a speech segment in a speech segment table according to the embodiment of the present invention is created from a temporally forward predicted speech waveform estimated from a spectral envelope parameter and a temporally backward predicted speech waveform.

FIG. 8 is a block diagram showing a configuration of a voice compression / decompression device according to the embodiment of the present invention.

[Explanation of symbols]

Reference Signs List 1 voice input unit 2 voice segment extraction unit 3 similarity determination unit 4 voice segment table 5 voice segment selection unit 6 encoding unit 7 encoded data output unit 8 spectrum envelope parameter extraction unit 9 expansion unit 10 residual generation unit 11 speech fragment Update unit h1 Speech segment to be processed A1, A2, A3, A4 Speech segment stored in speech segment table p1 Position of similar partial speech in speech segment

Claims (12)

[Claims] 1. A speech segment in a predetermined section is cut out from an input speech as a speech segment to be processed, a speech segment table including a plurality of types of speech segment groups is referred to, and each speech segment in the speech segment table and the speech to be processed are referred to. A process of comparing the similarity with the segment, selecting a speech segment with the highest similarity, and encoding the processing target speech segment based on data on the selected speech segment to generate encoded data. A voice compression / decompression method comprising: 2. After the encoded data is created, the encoded data is expanded, the expanded data is subtracted from the processing target speech piece to obtain a residual, and the residual waveform is Referring to a speech unit table including a plurality of types of speech unit groups, a process of comparing the similarity between each speech unit in the speech unit table and the residual waveform is performed one or more times to obtain encoded data. 2. The audio compression / decompression method according to claim 1, wherein: 3. A speech segment stored in the speech segment table is a speech segment created using an already-compressed and expanded speech waveform temporally behind the speech segment to be processed, and a spectral envelope parameter. It has at least a speech fragment created using the estimated temporally predicted speech waveform and the temporally backward predicted speech waveform, and a speech fragment created with a noise component. 3. The audio compression / expansion method according to claim 1, wherein the content is updated after the expansion process or after the extraction of the spectrum envelope parameter. 4. Each of the voice segments has a section longer in time than the voice segment to be processed, and when determining the similarity with the voice segment to be processed, processing is performed within a range of the length of each voice segment. 4. The speech compression / expansion method according to claim 1, wherein a similarity to the target speech piece is determined, and a speech piece having a portion having the highest similarity is selected. 5. The coded data is a voice segment number having a portion having the highest similarity, position data indicating which portion in the voice segment, and data represented by an amplitude adjustment parameter. 5. The audio compression / decompression method according to claim 4, wherein the data is data to which a spectrum envelope parameter is added as necessary. 6. A speech segment extraction unit for extracting a speech segment of a predetermined section set in advance from an input speech as a speech segment to be processed, a spectrum envelope parameter extraction unit for extracting a spectrum envelope parameter from the input speech, A voice segment table storing the segments, and a similarity determination unit that refers to the voice segment table, compares the similarities of the respective voice segments in the voice segment table and the similarity of the processing target voice segment, and determines the similarity. A voice segment selecting unit for selecting a voice segment having the highest similarity based on the similarity determined by the similarity determining unit; and An encoding unit that encodes the fragment, and outputs the data encoded by the encoding unit as encoded data, and, in some cases, the data encoded by the encoding unit. And a coded data output unit that generates and outputs coded data obtained by adding the spectrum envelope parameter extracted by the spectrum envelope parameter extraction unit to the data, as a constituent element. 7. A decompression unit for decompressing the data encoded by the encoding unit, and a residual generation unit for subtracting the data decompressed by the decompression unit from the speech unit to be processed to obtain a residual. A speech unit updating unit that updates the contents of the speech unit stored in the speech unit table using the data expanded by the expansion unit or the spectrum envelope parameter extracted by the spectrum envelope parameter extraction unit. 7. The audio compression / decompression device according to claim 6, comprising: 8. The similarity determination unit, speech unit selection unit, encoding unit, decompression unit, and residual generation unit form a loop in a processing procedure, and perform similarity judgment, speech unit selection, encoding, decompression, For the residual waveform obtained by performing the residual generation processing, referring to the speech piece table, a process of comparing the similarity between each speech piece in the speech piece table and the residual waveform, 1 8. The audio compression / decompression device according to claim 7, wherein the encoded data is created and output after performing the operations more than once. 9. A speech segment stored in the speech segment table is a speech segment created using an already-compressed and decompressed speech waveform temporally behind the speech segment to be processed, and a spectral envelope parameter. Estimated temporal forward predicted audio waveform and temporal backward predicted audio waveform, a speech segment created using the speech component, at least a speech segment created by a noise component, each speech segment by the speech update processing unit 7. The content is updated after decompression processing or after extraction of a spectrum envelope parameter.
9. The audio compression / decompression device according to any one of items 1 to 8. 10. Each of the speech segments has a longer section than the speech segment to be processed, and when determining the degree of similarity with the speech segment to be processed, the speech segment is processed within the length of each speech segment. The speech compression / expansion apparatus according to any one of claims 6 to 9, wherein a similarity to the speech piece is determined, and a speech piece having a portion having the highest similarity is selected. 11. The coded data is a voice segment number having the highest similarity portion, position data indicating which portion is in the voice segment, and data represented by an amplitude adjustment parameter. 11. The audio compression / decompression device according to claim 10, wherein the data is data to which a spectral envelope parameter is added as necessary. 12. A storage medium storing a voice compression / expansion processing program, the voice compression / expansion processing program cuts out a voice segment in a predetermined section from an input voice as a processing target voice segment, and includes a voice segment including a plurality of types of voice segment groups. Refer to the table,
The similarity between each speech segment in the speech segment table and the speech segment to be processed is compared, the speech segment having the highest similarity is selected, and the speech segment to be processed is selected based on the data about the selected speech segment. The voice segment is encoded and, if necessary, a process of creating encoded data to which a spectrum envelope parameter is added, and after the decompression process of the encoded data or after the extraction of the spectrum envelope parameter, A storage medium storing a voice compression / decompression processing program, wherein the content of each voice piece stored in a table is updated.